· Cybersecurity news

AI companion check: griefbots, the modern Ouija board — 5 security questions before you talk to a digital ghost

An AI griefbot promises the impossible: one more conversation with someone who is gone. A researcher who built a griefbot of herself just revealed what unsettled her most about it. Before grief, money, and your loved one's voice get mixed together, here are the five questions security and privacy demand you ask first.

The hook: a researcher immortalized herself — and got spooked

On October 16, 2026, UNILAD Tech reported on Dr Jenny Kidd of Cardiff University, who leads a research project called Synthetic Pasts (with Dr Eva Nieto McAvoy of King's College London) studying AI "deathbots" — systems that simulate a deceased person's wording, characteristics, and personality so the living can communicate with them. As part of the work, Kidd built a griefbot of herself.

What unsettled her was not a strong sense of presence. The systems her team built were poor at coherence: they contradicted themselves, lost track of ideas, and generated responses that didn't resemble her. What unsettled her was that a small amount of personal data could still produce something superficially resembling a person while being clearly inaccurate — mimicking stylistic patterns while missing the actual human being. She described the result as a "patterned caricature": a bland line of best fit, safe, smooth, and ultimately weird.

That finding is the security story in miniature: these systems don't need much of you to fake you.

What is an AI companion (griefbot)?

What is an AI griefbot? An AI griefbot — also called a deathbot or deadbot — is an AI system designed to simulate the voice, speech patterns, and personality of someone who has died. It trains on that person's digital traces: text messages, emails, social media posts, voice recordings, and photos. The result can be a text chatbot, a voice assistant, or an animated avatar that appears to speak from beyond the grave.

The industry is real — and already commercial

Most platforms run on subscriptions or freemium tiers. As one report on the field put it, memory becomes a service to design, measure, and sell: digital traces keep generating profit long after life ends.

5 security and privacy questions to ask first

1. Consent: who authorized this replica?

Some services let people pre-record themselves — consent is clear. But many griefbots are built by grieving relatives from the deceased's messages and photos, and the person being simulated never agreed to be simulated. There is no universal consent framework for post-mortem digital replicas. In one striking case, the family of a road-rage victim created an AI rendering of him to deliver a victim-impact statement in court — the avatar spoke words written by his sister, not generated by a model. It was arguably the first digitally simulated person in a legal proceeding, and it raises the same question in its sharpest form: who gets to speak for the dead?

2. Data: who holds the training material — and can it ever be deleted?

A griefbot trains on the most intimate dataset imaginable: private messages, voicemails, photos, emails. Before paying, demand answers: who stores this data, in which jurisdiction, can it be fully deleted on request — and what happens to it if the company folds? That last one is not hypothetical: HereAfter AI and Project December both shut down. When a grief-tech company dies, its customers' dead do not automatically get buried with it. Read the data-retention policy like a contract, because it is one.

3. Voice cloning: how little audio does it take to fake a loved one?

Modern voice cloning needs only seconds to minutes of audio to produce a convincing replica. A griefbot's familiar voice is the same technology behind voice-clone scams — and the bereaved are the most emotionally manipulable users on earth. Any platform that can clone your loved one's voice can, in principle, clone yours. Treat every voice interaction with a griefbot the way you would treat an unexpected call from a "relative": verify through a second channel before acting on anything it says. (See our AI voicemail detector and deepfake defense kit.)

4. Who is actually behind the bot — retrieval or generation?

Some systems are retrieval-driven: they surface things the person actually said. Others are LLM-powered: they invent new utterances "in the voice" of the dead. Kidd's finding matters here — generative systems contradict themselves and confabulate while wearing a loved one's face. A grieving user cannot easily tell which mode they're in. Ask the vendor, in writing: does this system ever generate statements the person never made? If the answer is evasive, you have your answer.

5. Cost traps: is grief being sold back to you monthly?

Freemium tiers, subscriptions, per-message pricing — the commercial griefbot model monetizes ongoing attachment. That creates a structural conflict: the company's revenue grows when you don't move on. Before subscribing, check the cancellation terms, whether a free tier locks your data behind a paywall, and what a year of "one more conversation" actually costs. Grief is not a subscription service; be wary of any product designed as one.

What the psychologists warn (reported)

Press and researchers report that psychologists caution interacting with an AI avatar may disrupt the natural grieving process — tethering people to the past rather than helping them move forward, prolonging denial instead of building acceptance. The researchers' own conclusion is blunter: a digital resurrection risks replacing the finality of loss with the endless availability of simulation, where the dead are always present, interactive, and updated. The evidence base for therapeutic benefit is thin; bold comfort claims deserve skepticism.

If you still want to: an honest-scope checklist

We are not here to tell you not to grieve your way. If a griefbot genuinely comforts you, that is your call. But go in with eyes open:

Honest scope: this page is a security and privacy briefing, not grief counseling and not legal advice. Consent law for digital replicas varies by jurisdiction and is actively evolving. If you are grieving, consider speaking with a qualified counselor.

Self-check: are you griefbot-ready?

1. A company offers to build a griefbot of your late father from his old messages. He never discussed it. You should…

2. The bot says something your loved one never said, in their voice. This is most likely…

3. Before uploading voicemails and photos, the most important document to read is…

4. A griefbot subscription keeps getting more expensive the longer you use it. This is…

5. The strongest protection against voice-clone manipulation via a griefbot is…

Frequently asked questions

What is a griefbot?

A griefbot (also called a deathbot or deadbot) is an AI system that simulates the wording, voice, and personality of a deceased person so the living can interact with them. It is trained on the person's digital traces: text messages, emails, social media posts, voice recordings, and photos.

Are AI griefbots accurate to the real person?

Not really. Cardiff University researcher Dr Jenny Kidd built a griefbot of herself and found it contradicted itself, lost track of ideas, and produced a shallow "patterned caricature" — mimicking stylistic quirks while missing the actual person. A small amount of data can produce something superficially convincing but clearly inaccurate.

Who can legally create a griefbot of someone who died?

The law is unsettled. Some services let people pre-record themselves with consent; others let grieving relatives build a replica of someone who never agreed. There is no universal consent framework for post-mortem digital replicas, which is one of the biggest unresolved questions in the field.

Can I delete a griefbot and its training data?

It depends on the platform's policy — and policies vary. Ask before you pay: who holds the data, can it be fully deleted on request, and what happens to it if the company shuts down. At least two early grief-tech companies (HereAfter AI and Project December) have already closed.

Do griefbots help or harm the grieving process?

Researchers are divided. Some users report comfort; psychologists warn that an always-available simulation can disrupt the natural grieving process and tether people to the past instead of helping them move forward. The evidence base is still thin — treat bold therapeutic claims with skepticism.

Related checks

Sources

Chat with Ingenious AI on WhatsApp →